Spacy: on_match functions with DependencyMatcher

[prilopes]

How to reproduce the behaviour

I'm trying to create a method to use on_match in DependencyMatcher. I've done this successfully for Matcher and follow much of the same logic.

When I get the matches using something like this:

matches = dep_matcher(doc)
for match in matches:
     print(match)

If I understand correctly match is a tuple with match_id (the rule that was triggered) and a list of nodes that match my pattern.

When defining an on_match method, I would expect matches[i] to return the same tuple, but that is not the case. Am I missing something??

I created the following script to help illustrate the problem

from spacy.matcher import DependencyMatcher
from spacy import load


def my_on_match(matcher, doc, i, matches):
    print('inside my_on_match')
    match = matches[i]
    print(match)
    print('================')


nlp = load('en_core_web_sm')
doc = nlp('The quick brown fox jumped over the lazy dog.')

dep_matcher = DependencyMatcher(nlp.vocab)

dep_pattern = [
    {"SPEC": {"NODE_NAME": "jumped"}, "PATTERN": {"ORTH": "jumped"}},
    {"SPEC": {"NODE_NAME": "fox", "NBOR_RELOP": ">", "NBOR_NAME": "jumped"}, "PATTERN": {"ORTH": "fox"}},
    {"SPEC": {"NODE_NAME": "quick", "NBOR_RELOP": ".", "NBOR_NAME": "jumped"}, "PATTERN": {"ORTH": "fox"}},
]

dep_matcher.add('no_on_match', None, dep_pattern)
dep_matcher.add('with_on_match', my_on_match, dep_pattern)

matches = dep_matcher(doc)

print('all matches')
for match in matches:
     print(match)

Your Environment

• spaCy version: 2.1.8
• Platform: Linux-4.15.0-66-generic-x86_64-with-Ubuntu-18.04-bionic
• Python version: 3.6.7

I might have found the problem

I was reading the code for DependencyMatcher and found these lines on dependencymatcher.pyx (start at line 240 on master branch and line 237 for tag 2.1.8):

    for i, (ent_id, nodes) in enumerate(matched_key_trees):
                on_match = self._callbacks.get(ent_id)
                if on_match is not None:
                    on_match(self, doc, i, matches)

So, if my line of thought is correct, on the last line of the excerpt (243 on master branch), should it be matched_key_trees instead of matches?

I'd appreciate any light on this issue. :-)

[adrianeboyd]

That looks like the correct analysis! (We definitely need to add documentation and more tests for this feature.)

[lock[bot]]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.